Wisp Library

wisp/ is the main folder for the wisp library. The folders can be divided up into framework (core components and functions to make everything work) and building blocks.

Framework

core contains core primitives and structs that are used everywhere throughout the code. The Rays class and the RenderBuffer are particularly important classes to be aware of.

csrc contains C++ and CUDA kernels which are used in the code.

framework contains state objects and event handlers to interface the core library with apps (like the renderer).

ops contains useful helper functions.

renderer contains everything that is used to build the interactive visualizer.

config contains code for the configuration system: a parser for yaml configs and the command-line interface (CLI) arguments.

Building Blocks

accelstructs contains classes that can be used as spatial acceleration structures for fast raytracing and query.

datasets contains classes for PyTorch datasets.

models contains various objects like MLPs and grids which have parameters that can be optimized.

tracers contains classes that can be used as forward maps, which are functions which map neural fields onto another domain. Currently this mostly contains mechanisms of tracing rays against neural fields.

trainers contains classes that are designed per tasks to actually train, log, save different workloads.

The Full Monty

The architecture figure describes how the major Wisp components relate to each other:

See also further README files accompanying the various wisp folders.